1 Big Data: Challenges and Opportunities Roberto V. Zicari Goethe University Frankfurt
2 This is Big Data. Every day, 2.5 quintillion bytes of data are created. This data comes from digital pictures, videos, posts to social media sites, intelligent sensors, purchase transaction records, cell phone GPS signals to name a few.
3 Big Data:The story as it is told from the Business Perspective. Big Data: The next frontier for innovation, competition, and productivity (McKinsey Global Institute) Data is the new gold : Open Data Initiative, European Commission (aim at opening up Public Sector Information).
4 Big Data: A Possible Definition Big Data refers to datasets whose size is beyond the ability of typical database software tools to capture, store, manage and analyze (McKinsey Global Institute) This definition is Not defined in terms of data size (data sets will increase) Vary by sectors (ranging from a few dozen terabytes to multiple petabytes) (1petabyte is 1,000 terabytes (TB)
5 Where is Big Data? (Big) Data is in every industry and business function and are important factor for production (McKinsey Global Institute) - (estimated 7 exabytes of new data enterprises globally stored in MGI)
6 What is Big Data supposed to create? Value (McKinsey Global Institute): Creating transparencies Discovering needs, expose variability, improve performance Segmenting customers Replacing/supporting human decision making with automated algorithms Innovating new business models,products,services
7 How Big Data will be used? Key basis of competition and growth for individual firms (McKinsey Global Institute). E.g. retailer embracing big data has the potential to increase its operating margin by more than 60 percent.
8 How to measure the value of Big Data? Consider only those actions that essentially depends on the use of big data. (McKinsey Global Institute)
9 Big Data can generate financial value across sectors Health care Public sector administration Global personal location data Retail Manufacturing (McKinsey Global Institute)
10 Limitations Shortage of talent necessary for organizations to take advantage of big data. Knowledge in statistics and machine learning, data mining. Managers and Analysts who make decision by using insights from big data. (McKinsey Global Institute)
11 Data Policies Issues (McKinsey Global Institute) e.g. storage, computing, analytical software e.g.new types of analyses Technology and techniques e.g. Privacy, security, intellectual property, liability Access to Data e.g. integrate multiple data sources Industry structure e.g. lack of competitive pressure in public sector
12 Big Data: Challenges Data, Process, Management Data: Volume (dealing with the size of it) In the year 2000, 800,000 petabytes (PB) of data stored in the world (source IBM). Expect to reach 35 zettabytes (ZB) by Twitter generates 7+ terabytes (TB) of data every day. Facebook 10TB. Variety (handling multiplicity of types, sources and formats) Sensors, smart devices, social collaboration technologies. Data is not only structured, but raw, semi structured, unstructured data from web pages, web log files (click stream data), search indexes, s, documents, sensor data, etc.
13 Challenges cont. Data: Data availability is there data available, at all? A good process will, typically, make bad decisions if based upon bad data. Data quality how good is the data? How broad is the coverage? How fine is the sampling resolution? How timely are the readings? How well understood are the sampling biases? e.g. what are the implications in, for example, a Tsunami that affects several Pacific Rim countries? If data is of high quality in one country, and poorer in another, does the Aid response skew unfairly toward the well-surveyed country or toward the educated guesses being made for the poorly surveyed one? (Paul Miller)
14 Challenges Data Velocity (reacting to the flood of information in the time required by the application) Stream computing: e.g. Show me all people who are currently living in the Bay Area flood zone - continuosly updated by GPS data in real time. (IBM) Veracity (how can we cope with uncertainty, imprecision, missing values, mis-statements or untruths?) Data discovery is a huge challenge (how to find high-quality data from the vast collections of data that are out there on the Web). Determining the quality of data sets and relevance to particular issues (i.e., is the data set making some underlying assumption that renders it biased or not informative for a particular question). Combining multiple data sets
15 Challenges cont. Data Data comprehensiveness are there areas without coverage? What are the implications? Personally Identifiable Information much of this information is about people. Can we extract enough information to help people without extracting so much as to compromise their privacy? Partly, this calls for effective industrial practices. Partly, it calls for effective oversight by Government. Partly perhaps mostly it requires a realistic reconsideration of what privacy really means. (Paul Miller)
16 Challenges cont. Data: Data dogmatism analysis of big data can offer quite remarkable insights, but we must be wary of becoming too beholden to the numbers. Domain experts and common sense must continue to play a role. e.g. It would be worrying if the healthcare sector only responded to flu outbreaks when Google Flu Trends told them to. (Paul Miller)
17 Challenges cont. Process The challenges with deriving insight include - capturing data, - aligning data from different sources (e.g., resolving when two objects are the same), - transforming the data into a form suitable for analysis, - modeling it, whether mathematically, or through some form of simulation, - understanding the output visualizing and sharing the results, (Laura Haas)
18 Challenges cont. Management: data privacy, security, and governance: - ensuring that data is used correctly (abiding by its intended uses and relevant laws), - tracking how the data is used, transformed, derived, etc, - and managing its lifecycle. Many data warehouses contain sensitive data such as personal data. There are legal and ethical concerns with accessing such data. So the data must be secured and access controlled as well as logged for audits (Michael Blaha).
19 Let`s take some time to critically review this story.
20 Examples of BIG DATA USE CASES Log Analytics Fraud Detection Social Media and Sentiment Analysis Risk modeling and management Energy sector
21 Big Data: The story as it is told from the Technology Perspective. What are the main technical challenges for big data analytics? In the Big Data era the old paradigm of shipping data to the application isn`t working any more. Rather, the application logic must come to the data or else things will break: this is counter to conventional wisdom and the established notion of strata within the database stack. With terabytes, things are actually pretty simple -- most conventional databases scale to terabytes these days. However, try to scale to petabytes and it`s a whole different ball game. (Florian Waas) Confirms Gray`s Laws of Data Engineering: Take the Analysis to the Data!
22 Seamless integration Instead of stand-alone products for ETL, BI/reporting and analytics we have to think about seamless integration: in what ways can we open up a data processing platform to enable applications to get closer? What language interfaces, but also what resource management facilities can we offer? And so on. (Florian Waas)
23 Scale and performance requirements strain conventional databases. The problems are a matter of the underlying architecture. If not built for scale from the ground-up a database will ultimately hit the wall -- this is what makes it so difficult for the established vendors to play in this space because you cannot simply retrofit a 20+ year-old architecture to become a distributed MPP database over night. (Florian Waas)
24 Big Data Analytics In the old world of data analysis you knew exactly which questions you wanted to asked, which drove a very predictable collection and storage model. In the new world of data analysis your questions are going to evolve and change over time and as such you need to be able to collect, store and analyze data without being constrained by resources. Werner Vogels, CTO, Amazon.com
25 How to analyze? It can take significant exploration to find the right model for analysis, and the ability to iterate very quickly and fail fast through many (possible throwaway) models -at scale - is critical. (Shilpa Lawande)
26 Faster As businesses get more value out of analytics, it creates a success problem - they want the data available faster, or in other words, want real-time analytics. And they want more people to have access to it, or in other words, high user volumes. (Shilpa Lawande)
27 Semi-structured Web data. A/B testing, sessionization, bot detection, and pathing analysis all require powerful analytics on many petabytes of semi-structured Web data.
28 Big Data Analytics In order to analyze Big Data, the current state of the art is a parallel database or NoSQL data store, with a Hadoop connector. Concerns about performance issues arising with the transfer of large amounts of data between the two systems. The use of connectors could introduce delays, data silos, increase TCO.
29 Scalability Scalability has three aspects: data volume, hardware size, and concurrency.
30 Which Analytics Platform for Big Data? Big Data in the Database World (early 1980s till now): - Parallel Data Bases. Shared-nothing architecture, declarative set-oriented nature of relational queries, divide and conquer parallelism (e.g. Teradata) - Re-implemention of relational databases (e.g. HP/Vertica, IBM/Netezza, Teradata/ Aster Data, EMC/ Greenplum.) Big Data in the Systems World (late 1990s) - Apache Hadoop (inspired by Google GFS, MapReduce), (contributed by large Web companies.e.g. Yahoo!, Facebook - Google BigTable, - Amazon Dynamo
31 Parallel database software stack (Michael J. Carey) SQL-->SQL Compiler Relational Dataflow layer (runs the query plans, orchestrate the local storage managers, deliver partitioned, shared-nothing storage services for large relational tables) Row/Column Storage Manager (record-oriented: made up of a set of row-oriented or column oriented storage managers per machine in a cluster) No open-source parallel database exists! SQL is the only way into the system architecture. Monolithic: Can`t safely cut into them to access inner functionalities
32 Hadoop software stack (Michael J. Carey) HiveQL. PigLatin, Jaql script-->hiveql/pig/jaql (High-level languages) Hadoop M/R job-->hadoop Map Reduce Dataflow Layer/ (for batch analytics, applies Map ops to the data in partitions of an HDFS file, sorts, and redistributes the results based on key values in the output data, then performs reduce on the groups of output data items with matching keys from the map phase of the job). Get/Put ops-->hbase Key-value Store (accessed directly by client app or via Hadoop for anylstics needs) Hadoop Distributed File System (byte oriented file abstraction- files appears as a very large contiguos and randomly addressible sequence of bytes
33 Hadoop Pros: open source Hadoop Pros (Michael J. Carey) non-monolithic support for access to file-based external data support for automatic and incremental forward-recovery of jobs with failed task ability ot schedule very large jobs in smaller chunks automatic data placement and rebalancing as data grows and machines come and go. support for replication and machine fail-over without operation intervention
34 Hadoop Cons (Michael J. Carey) Hadoop Cons: - questionable to layer a record-oriented data abstraction on top of a giant globally-sequenced byte-stream file abstraction. (e.g. HDFS is unaware of record boundaries. broken records instead of fixed-lenght file splits, i.e. a record with some of its bytes in one spit and some inn the next) - questionable building a parallel data runtime on top of a unary operator model (map, reduce, combine). E.g. Performing joins with MapReduce. - questionable building a key-value store layer with a remote query access at the next layer. Pushing queries down to data is likely to outperform pulling data up to queries. - lack of schema information, today is flexible, but a recipe for future difficulties. E.g. Future maintainers of applications will likely have problems in fixing bugs related to changes or assumptions about the structure of data files in HDFS. (This was one of the very early lessons in the DB world). - Not addressed single system performance, focusing solely on scaleout.
35 Big Data stack Many academics are being too shy about questioning and rethinking the Big Data stack forming around Hadoop today, perhaps because they are embarassed about having fallen asleep in the mid-1990s (with respect to parallel data issues) and waking up in the world as it exists today (Michael Carey, EDBT keynote 2012)
36 Big Data stack Rather that trying to modify the Hadoop code base to add indexing or data co-clustering support, or gluing open-source database systems underneath Hadoop`s data input APIs, we believe that database researchers should be asking Why? or What if we`d meant to design an open software stack with records at the bottom and a strong possibility of a higher-level language API at the top? (Michael Carey, EDBT keynote 2012)
37 Two Research Projects The ASTERIX project (UC Irvinestarted 2009) open-source Apache-style licence. The Stratosphere project (TU Berlin)
38 Apache Hadoop Apache Hadoop provides a new platform to analyze and process Big Data. Hadoop was inspired by Google`s MapReduce and Google File System (GFS) papers.
39 Hadoop is really an ecosystems of projects Higher-level declarative languages for writing queries and data analysis pipelines Pig (Yahoo!) - relational-like algebra (60% of Yahoo! MapReduce use cases) PigLatin Hive (Facebook) also inspired by SQL (90% of Facebook MapReduce use cases) Jaql (IBM) Load Transform Dump and store More Flume Zookeeper Hbase Oozie Lucene Avro Etc.
40 Apache Hadoop benefits The combination of scale, ability to process unstructured data along with the availability of machine learning algorithms and recommendation engines creates the opportunity to build new game changing applications.
41 Hadoop Limitations Hadoop can give powerful analysis, but it is fundamentally a batch-oriented paradigm. The missing piece of the Hadoop puzzle is accounting for real time changes.
42 Hadoop Limitations e.g. HDS has a centralized metadata store (NameNode), which represents a single point of failure without availability. When the NameNode is recovered, it can take a long time to get the Hadoop cluster running again. Difficult to use Work is in progress to fix this from vendors of commercial Hadoop distributions (e.g. MapR, etc.) by re-implementing Hadoop components.
43 Big Data Dichotomy Analytics: MapReduce, Hadoop Developers of very large scale user-facing Web sites implemented key-value stores Google Big Table Amazon Dynamo Apache Hbase (open source BigTable clone), Apache Cassandra (open source Dynamo clone),
44 Hadoop By not requiring a schema first, Hadoop provides a great tool for exploratory analysis of the data, as long as you have the software development expertise to write Map Reduce programs. Hadoop assumes that the workload it runs will be long running, so it makes heavy use of checkpointing at intermediate stages. This means parts of a job can fail, be restarted and eventually complete successfully. There are no transactional guarantees. (Shilpa Lawande)
45 Why Using Hadoop? We chose Hadoop for several reasons. First, it is the only available framework that could scale to process 100s or even 1000s of terabytes of data and scale to installations of up to 4000 nodes. Second, Hadoop is open source and we can innovate on top of the framework and inside it to help our customers develop more performant applications quicker. Third, we recognized that Hadoop was gaining substantial popularity in the industry with multiple customers using Hadoop and many vendors innovating on top of Hadoop. Three years later we believe we made the right choice. We also see that existing BI vendors such as Microstrategy are willing to work with us and integrate their solutions on top of Elastic. MapReduce. (Werner Vogels, VP and CTO Amazon)
46 Vertica (NewSQL) and Hadoop Vertica (*) and Hadoop are both systems that can store and analyze large amounts of data on commodity hardware. The main differences are how the data gets in and out, how fast the system can perform, and what transaction guarantees are provided. Also, from the standpoint of data access, Vertica`s interface is SQL and data must be designed and loaded into a SQL schema for analysis. With Hadoop, data is loaded AS IS into a distributed file system and accessed programmatically by writing Map-Reduce programs. ( Shilpa Lawande) (*) columnar database engine including sorted columnar storage, a query optimizer and an execution engine, provides standard ACID transaction semantics on loads and queries
47 Analytics at ebay: technical challenges Main technical challenges for big data analytics at ebay: I/O bandwidth: limited due to configuration of the nodes. Concurrency/workload management: Workload management tools usually manage the limited resource. For many years EDW systems bottle neck on the CPU; big systems are configured with ample CPU making I/O the bottleneck. Vendors are starting to put mechanisms in place to manage I/O, but it will take some time to get to the same level of sophistication. Data movement (loads, initial loads, backup/restores): As new platforms are emerging you need to make data available on more systems challenging networks, movement tools and support to ensure scalable operations that maintain data consistency (Tom Fastner)
48 Analytics at ebay: Platforms for Analytics 3 different platforms for Analytics: A) EDW: Dual systems for transactional (structured) data; Teradata 3.5PB and 2.5 PB spinning disk; 10+ years experience; very high concurrency; good accessibility; hundreds of applications. B) B) Singularity: deep Teradata system for semistructured data; 36 PB spinning disk; lower concurrency that EDW, but can store more data; biggest use case is User Behavior Analysis; largest table is 1.2 PB with ~1.9 Trillion rows. C) C) Hadoop: for unstructured/complex data; ~40 PB spinning disk; text analytics, machine learning; has the User Behavior data and selected EDW tables; lower concurrency and utilization. (Tom Fastner))
49 Analytics at ebay: Scalability and Performance DW: We model for the unknown (close to 3rd NF) to provide a solid physical data model suitable for many applications, that limits the number of physical copies needed to satisfy specific application requirements. A lot of scalability and performance is built into the database, but as any shared resource it does require an excellent operations team to fully leverage the capabilities of the platform Singularity: The platform is identical to EDW, the only exception are limitations in the workload management due to configuration choices. But since we are leveraging the latest database release we are exploring ways to adopt new storage and processing patterns. Some new data sources are stored in a denormalized form significantly simplifying data modeling and ETL. On top we developed functions to support the analysis of the semi-structured data. It also enables more sophisticated algorithms that would be very hard, inefficient or impossible to implement with pure SQL. One example is the pathing of user sessions. However the size of the data requires us to focus more on best practices (develop on small subsets, use 1% sample; process by day), Hadoop: The emphasis on Hadoop is on optimizing for access. The reusability of data structures (besides raw data) is very low. (Tom Fastner)
50 Analytics at ebay: Un-structured data Un-structured data is handled on Hadoop only. The data is copied from the source systems into HDFS for further processing. We do not store any of that on the Singularity (Teradata) system (Tom Fastner)
51 Analytics at ebay: Use of Data management technologies ETL: AbInitio, home grown parallel Ingest system. Scheduling: UC4. Repositories: Teradata EDW; Teradata Deep system; Hadoop. BI: Microstrategy, SAS, Tableau, Excel. Data modeling: Power Designer. Adhoc: Teradata SQL Assistant; Hadoop Pig and Hive. Content Management: Joomla based. (Tom Fastner)
52 Analytics at ebay: Cloud computing and open source We do leverage internal cloud functions for Hadoop; no cloud for Teradata. Open source: committers for Hadoop and Joomla; strong commitment to improve those technologies (Tom Fastner)
53 Analytics at ebay: use of analytics Ebay is rapidly changing, and analytics is driving many key initiatives like buyer experience, search optimization, buyer protection or mobile commerce. We are investing heavily in new technologies and approaches to leverage new data sources to drive innovation. (Tom Fastner)
54 Hadoop users Advanced users of Hadoop are looking to go beyond batch uses of Hadoop to support realtime streaming of content. How many advanced users? New users need Hadoop to become easier. Need it to be easier to develop Hadoop applications, deploy them and run them in a production environment. - Is there a real need for it?
55 Applicability of Hadoop Promises from Hadoop vendors: Product recommendations, ad placements, customer churn, patient outcome predictions, fraud detection and sentiment analysis are just a few examples that improve with real time information. Organizations are also looking to expand Hadoop use cases to include business critical, secure applications that easily integrate with file-based applications and products. With mainstream adoption comes the need for tools that don`t require specialized skills and programmers. New Hadoop developments must be simple for users to operate and to get data in and out. This includes direct access with standard protocols using existing tools and applications. --> See also Big Data Myth later
56 Hadoop distributions challenges Getting data in and out of Hadoop. Some Hadoop distributions are limited by the append-only nature of the Hadoop Distributed File System (HDFS) that requires programs to batch load and unload data into a cluster. Deploying Hadoop into mission critical business projects. The lack of reliability of current Hadoop software platforms is a major impediment for expansion. Protecting data against application and user errors. Hadoop has no backup and restore capabilities. Users have to contend with data loss or resort to very expensive solutions that reside outside the actual Hadoop cluster.
57 Hadoop and the Cloud In general people are concerned with the protection and security of their data. Hadoop in the cloud: Amazon has a significant web-services business around Hadoop What about traditional enterprises?
58 VoltDB and Hadoop VoltDB is not focused on analytics. We believe they should be run on a companion data warehouse. Most of the warehouse customers I talk to want to keep increasing large amounts of increasingly diverse history to run their analytics over. The major data warehouse players are routinely being asked to manage petabyte-sized data warehouses. VoltDB is intended for the OLTP portion, and some customers wish to run Hadoop as a data warehouse platform. To facilitate this architecture, VoltDB offers a Hadoop connector (Mike Stonebraker)
59 Couchbase (NoSQL) and Hadoop In some applications Couchbase (NoSQL) is used to enhance the batchbased Hadoop analysis with real time information, giving the effect of a continuous process.
60 Couchbase (NoSQL) and Hadoop Hot data lives in Couchbase in RAM. Essentially move the data out of Couchbase into Hadoop when it cools off. Connector to Apache Sqoop (Top-Level Apache project since March of 2012): a tool designed for efficiently transferring bulk data between Haddop and relational databases.
61 NoSQL and Hadoop In my opinion the primary interface will be via the real time store, and the Hadoop layer will become a commodity. That is why there is so much competition for the NoSQL brass ring right now. --J. Chris Anderson.
62 Benchmarking SQL and NoSQL data stores There is a scarcity of benchmarks to substantiate the many claims made of scalabilty of NoSQL vendors. NoSQL data stores do not qualify for the TPC-C benchmark, since they relax ACID transaction properties. How can you then measure and compare the performance of the various NoSQL data stores instead?
63 Yahoo! YCSB benchmark Vendors are making a lot of claims about latency, throughput and scalability without much proof, Yahoo YCSB benchmark is source of good comparisons.
64 Benchmark for Cloud Serving Systems A team of researchers composed of Adam Silberstein, Brian F. Cooper, Raghu Ramakrishnan, Russell Sears, and Erwin Tam, all at Yahoo! Research Silicon Valley, developed a new benchmark for Cloud Serving Systems, called YCSB.
65 Measuring the scalability of SQL and NoSQL systems. They open-sourced the benchmark about a year ago. The YCSB benchmark appears to be the best to date for measuring the scalability of SQL and NoSQL systems.
66 NoSQL Performance There are many design decisions to make when building NoSQL systems, and those decisions have a huge impact on how the system performs for different workloads (e.g., read-heavy workloads vs. write-heavy workloads), how it scales, how it handles failures, ease of operation and tuning, etc.
67 YCSB vs. TPC-C At a high level, there is a lot in common with TPC-C and other OLTP benchmarks: Query latency and overall system throughput. BUT Queries are very different. TPC-C contains several diverse types of queries meant to mimic a company warehouse environment. Some queries execute transactions over multiple tables; some are more heavyweight than others. In contrast, the web applications YCSB is benchmarking tend to run a huge number of extremely simple queries.
68 YCSB vs. TPC-C Consider a table where each record holds a user`s profile information. Every query touches only a single record, likely either reading it, or reading+writing it. YCSB does include support for skewed workloads; some tables may have active sets accessed much more than others. Focused on simple queries. Ease of creating a new suite of benchmarks using the YCSB framework.
69 Parameters Performance: refers to the usual metrics of latency and throughput, with the ability to scale out by adding capacity. Elasticity: refers to the ability to add capacity to a running deployment on-demand, without manual intervention (e.g., to re-shard existing data across new servers).
70 Results and Lesson. -Result #1. We knew the systems made fundamental decisions to optimize writes or optimize reads. It was nice to see these decisions show up in the results.example: in a 50/50 workload, Cassandra was best on throughput. In a 95% read workload, PNUTS caught up and had the best latencies. -Result #2. The systems may advertise scalability and elasticity, but this is clearly a place where the implementations needed more work. Ref. elasticity experiment. Ref. HBase with only 1-2 nodes. -Lesson. We are in the early stages. The systems are moving fast enough that there is no clear guidance on how to tune each system for particular workloads. (Adam Silberstein, Raghu Ramakrishnan)
71 Availability The authors open-sourced the benchmark about a year ago. It is available at:
72 Big Data in Data Warehouse or in Hadoop? Data Warehouse: Structured data, Data trusted Hadoop: Semistructured and unstructured data. Data not trusted. Work in progress to develop tools
73 How easy is Hadoop? There are only a few Facebook-sized IT organizations that can have 60 Stanford PhDs on staff to run their Hadoop infrastructure. The others need it to be easier to develop Hadoop applications, deploy them and run them in a production environment. -- John Schroeder.
74 Big Data Search There is no single set formula for extracting value from big data; it will depend on the application. There are many applications where simply being able to comb through large volumes of complex data from multiple sources via interactive queries can give organizations new insights about their products,customers, services, etc. Being able to combine these interactive data explorations with some analytics and visualization can produce new insights that would otherwise be hidden.
75 Enterprise Search Enterprise Search implies being able to search multiple types of data generated by an enterprise. Apache Solr. There`s an ecosystem of tools that build on Solr, Solr support or implementing a proprietary full-text search engine?
76 Big Data Search: Example For example, real-time co-occurrence analysis new insights about how products are being used. It was analysis of social media that revealed that Gatorade is closely associated with flu and fever, and with the ability to drill seamlessly from high-level aggregate data into the actual source social media posts shows that many people actually take Gatorade to treat flu symptoms. Geographic visualization shows that this phenomenon may be regional. (David Gorbert)
77 Big Data myth We believe that in-memory / NewSQL is likely to be the prevalent database model rather than NoSQL due to three key reasons: 1) the limited need of petabyte-scale data today even among the NoSQL deployment base, 2) very low proportion of databases in corporate deployment which requires more than tens of TB of data to be handles, and 3) lack of availability and high cost of highly skilled operators (often post-doctoral) to operate highly scalable NoSQL clusters. (Marc Geall, Research Analyst, Deutsche Bank AG/London)
78 to review this story Let`s take time
79 Big Data: The other story Very few people seem to look at how Big Data can be used for solving social problems. Most of the work in fact is not in this direction. Why this? What can be done in the international research/development community to make sure that some of the most brilliant ideas do have an impact also for social issues?